Skip to main content

A PyTorch dataset of the eyes found in the OmniArt dataset

Project description

OmniArt Eye Dataset

This package provides a dataset of 118,576 painted eyes. These eyes are extracted from about 245,000 paintings from the OmniArt dataset. The dataset provides the eyes as images, the colour of the iris, and metadata from the OmniArt dataset.

Usage

The dataset can be used like any other PyTorch dataset. It extends the ImageFolder class to provide the images and labels/colour and in addition attaches the OmniArt metadata as a dictionary. The following classes are used, and how many of that class exist in the dataset:

Colour Count
Amber 3,114
Blue 18,926
Brown 42,094
Gray 7,637
Grayscale 16,234
Green 5,220
Hazel 1,578
Irisless 18,282
Negative 96,042
Red 5,491

The negative class exists to be able to classify non-eye images. It contains samples of primarily noise and facial areas, such as closed eyelids.

Example

The dataset can be used in the following way

import torch
import matplotlib.pyplot as plt
import numpy as np
import torchvision.utils as vutils
from torchvision.transforms import transforms

from omniart_eye_dataset import OmniArtEyeDataset

dataset = OmniArtEyeDataset(transform=transforms.Compose([
                               transforms.Resize(50),
                               transforms.CenterCrop(50),
                               transforms.ToTensor(),
                           ]))

dataloader = torch.utils.data.DataLoader(dataset, batch_size=64, shuffle=True, num_workers=4)

# Take 64 random entries
images, color, metadata = next(iter(dataloader))

# Plot the entries
plt.figure(figsize=(10, 10))
plt.axis("off")
plt.title("OmniArt eyes")
plt.imshow(np.transpose(vutils.make_grid(images, padding=5, normalize=True), (1, 2, 0)))
plt.show()

Sample eyes

Related

This dataset has already been used to train a classifier and painted eye generator.

Project origin

This package is part of a Master's thesis at the University of Amsterdam.

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Files for omniart-eye-dataset, version 0.1.1
Filename, size File type Python version Upload date Hashes
Filename, size omniart_eye_dataset-0.1.1-py3-none-any.whl (3.4 kB) File type Wheel Python version py3 Upload date Hashes View

Supported by

Pingdom Pingdom Monitoring Google Google Object Storage and Download Analytics Sentry Sentry Error logging AWS AWS Cloud computing DataDog DataDog Monitoring Fastly Fastly CDN DigiCert DigiCert EV certificate StatusPage StatusPage Status page